WorksPhysics and Astronomy
Semi-analytical model for the rising sheet generated by droplet-pair impact
Abstract
When two low-Ohnesorge-number drops impact a dry substrate simultaneously, their spreading lamellae collide and lift a free-standing vertical sheet. The sheet grows by inertial feeding from the spreading drops and is pulled back by capillary retraction at its rim. We develop a semi-analytical model for this rising sheet by extending the single-drop impact description of~\citet{Gordillo2019} to the two-drop geometry. The thin-film flow in the sheet is coupled at its base to the colliding lamellae and at its apex to a capillary-retarded rim. The sheet interior is then solved along ballistic characteristics in two stages: a lamella-fed stage, for which the velocity and thickness fields can be obtained in closed form, and a post-lamella stage, for which the inlet conditions are taken from simulations. The resulting framework gives the three-dimensional velocity and thickness fields and therefore the full sheet shape. On the centreline, the apex height and local thickness are obtained explicitly, showing that the different Weber-number exponents reported in the literature arise from a crossover rather than from a single universal scaling law. At sufficiently large Weber number, the apex pinches off. A linear Rayleigh–Plateau analysis, using the time-dependent jet diameter and deceleration predicted by the model, then bounds the maximum attainable height and closes the description of the pinch-off regime.
Introduction
The impact of liquid drops on solid surfaces is a classical problem in fluid mechanics that has been studied extensively for decades because of its rich interfacial dynamics and broad relevance to applications [2, 3, 4].For an isolated drop, the sequence of events following impact, including radial spreading, lamella formation and ejection, rim growth, and eventual retraction or breakup, is now reasonably well understood through a combination of experiments, numerical simulations, and reduced theory [1, 2, 5, 6, 7, 8, 9, 10]. In many practical situations, however, drops do not impact in isolation but as part of dense sprays, in which neighboring impacts interact on comparable spatial and temporal scales [11, 12, 13]. Such interactions introduce collective dynamics that are absent from the single-drop problem. Yet, despite their practical importance, multi-drop impacts remain far less understood than the canonical case of a single impacting drop [14].
The simultaneous impact of two identical drops provides the simplest geometry in which such interaction effects appear. When the drops impact sufficiently close to each other, their spreading rims collide at the symmetry plane. The collision redirects part of the horizontal momentum upward and produces a free-standing sheet, hereafter referred to as the central rising sheet. The sheet is bounded at its upper edge by a retracting rim. This structure was described by Barnes et al. [15], who observed that it can destabilize and generate secondary droplets much larger than those produced by single-drop splashing. Subsequent studies have examined different aspects of this process. Roisman et al. [16] proposed an early model for the sheet height, but without direct experimental validation. Ersoy and Eslamian [17] characterized the semilunar shape of the sheet. More recently, Goswami and Hardalupas [18] measured the time-dependent sheet height, width, and thickness over a range of Weber numbers and drop spacings, and proposed an empirical scaling for the maximum height. Goswami and Hardalupas [19] subsequently extended this characterization over nearly two decades in liquid viscosity, corresponding to Ohnesorge numbers from 0.002 to 0.177. They reported a morphological transition from well-defined semilunar sheets to short-lived liquid bumps, with capillary waves and rim corrugations progressively suppressed as viscosity increased. Beyond these experimental studies, Zhang et al. [20] used three-dimensional direct numerical simulations to develop an energy-balance model for the maximum sheet height, in which the viscous dissipation was calibrated from the numerical data. Together, these studies have established the central rising sheet as the defining hydrodynamic feature of simultaneous two-drop impact. However, existing descriptions of its maximum height remain based on empirical scalings or numerical calibration. A predictive model for the unsteady sheet dynamics, linking the lamella collision to the sheet trajectory, thickness evolution, and rim motion, is still lacking.
Motivated by this gap, we develop a semi-analytical model for the unsteady central sheet. The model starts from the lamella description for single-drop spreading and extends it to the two-drop geometry. In § 0.2, we first summarize the elements of the single-drop lamella solution [1] that provide the inlet conditions for the sheet. In § III, we introduce the two-drop geometry, specify the matching conditions at the collision line and at the rim, and solve the three-dimensional thin-film problem by following the ballistic characteristics of fluid elements through the lamella-fed (L) and post-lamella (PL) stages. In the L stage, the explicit lamella solution from Gordillo et al. [1] provides the boundary condition for the rising sheet, whereas in the PL stage we employ an empirical boundary condition extracted from fully numerical simulations. The resulting sheet profiles are then compared with experiments and simulations. Since the apex height is the main observable, § 0.4 reduces the three-dimensional solution to the centreline section. This reduction gives a closed-form expression for the height during the lamella-fed stage and a continuation into the post-lamella stage, which are tested against measured and simulated sheet. The model predicts an increasing height with Weber number, but in experiments the sheet height is ultimately limited by capillary breakup. We therefore close the description in § 0.5 by introducing a capillary cutoff: the time-dependent jet diameter and deceleration predicted by the model are used as inputs to a linear Rayleigh–Plateau analysis of the apex. Conclusions and limitations are given in § VI.
Dynamics from a single impacting drop
Consider a droplet of dimensional radius , density , dynamic viscosity , and surface tension impacting a dry, flat substrate with normal velocity . The time denotes the onset of impact. Quantities with an asterisk are dimensional, whereas all other variables are nondimensionalized using for lengths, for velocities, for times, and for stress. The governing dimensionless parameters are the Weber number , the Reynolds number , and the Ohnesorge number . We focus on the inertia-dominated regime relevant to millimetric drop impacts, for which and . The low-viscosity regime considered here corresponds to , consistent with the parameter range treated by Gordillo et al. [1].
For the single-drop impact problem, we use cylindrical coordinates centered at the impact point. After contact, the liquid motion can be divided into three regions [1], as shown in Figure 1a: the drop region, ; the lamella region, ; and the rim region. The radius separating the drop and lamella regions is taken as . This expression was derived by Riboux and Gordillo [8] from Wagner’s linearized potential-flow theory [21]. Although the derivation is asymptotically valid for , comparisons with numerical lamella profiles show good agreement up to times of order unity [1, 22].
Figure 1. Schematic of droplet impact. (a) Single droplet impact: (top) an undeformed spherical drop of radius and material properties descending at velocity prior to impact, where denotes quantities with dimensions; (bottom) post-impact spreading on the substrate, comprising (i) the undeformed drop region , (ii) the lamella with depth-averaged radial velocity , and (iii) the rim of diameter located at and advancing at velocity . (b) Simultaneous pair-drop impact at dimensionless half-spacing , with the two spreading lamellae meeting along the mid-plane and ejecting a central rising sheet. (c) Cross-section of the central sheet at , the plane to which the present analysis is restricted, of apex height and local thickness , topped by a rim of diameter rising at velocity . (d) Front view of the central sheet (looking along ), illustrating its semilunar profile, the depth-averaged vertical velocity within the sheet, the lateral spreading velocity of the surrounding lamellae whose divergence at produces the sink term in the sheet continuity equation (19), and the collision front marking the lateral extent of the meeting lamellae along . Dashed lines trace the growth trajectory of a rim element from its origin on the collision front to its current position on the rim.
Within the lamella region, , the liquid spreads as a thin film with depthaveraged radial velocity and thickness . At the front, , surface tension collects the film into a toroidal rim. The governing equations for the lamella and rim, together with the solutions used here, are summarized in Appendix A following Gordillo
et al. [1]. These results provide the input for the pair-drop problem considered below.
The lamella model is based on three main approximations. First, pressure gradients within the lamella are neglected because the film is slender. Second, the outer lamella flow is treated as inviscid in the high-Reynolds-number limit. Third, the effect of the substrate enters through a thin viscous boundary layer, which provides the leading correction to the inviscid spreading flow. Accordingly, the depth-averaged velocity is expanded as , where is the leading inviscid solution and is the first correction due to boundary-layer friction at the wall. Once and are known, the rim velocity and rim radius are obtained from the rim mass and momentum balances, using the lamella values and at the front. For later convenience, we define , which measures the viscous attenuation of the radial lamella flow.
We retain this first-order correction because it has an explicit analytical form Eq. (39) and therefore allows the influence of wall-induced viscous attenuation on the subsequent sheet dynamics to be incorporated quantitatively. In particular, as shown in Eqs. (25) and (30), this correction modifies the predicted scaling of the maximum sheet height, leading to a dependence distinct from the inviscid scaling .
Three-dimensional dynamics of rising sheet
Geometry, notation and assumptions
We now consider the simultaneous impact of two identical drops on a dry substrate, with dimensional center-to-center separation , as shown in Figure 1b. Each drop has the same fluid properties and impact conditions as in §II. The separation is measured by the dimensionless half-spacing .
We use a Cartesian coordinate system centered at the midpoint between the two impact points. The axis is normal to the substrate, the axis connects the two impact centers, and the axis lies along the collision line. With this convention, the central sheet rises in the – plane and is thin in the direction. Throughout this section, lower-case symbols, such as , , , and , denote lamella quantities inherited from the single-drop problem in §II. Upper-case symbols are used for the central sheet: and are the depth-averaged velocity components in the and directions, is the sheet thickness, is the rim cross-sectional diameter, is the apex height, and is the collision front. These quantities are indicated in Figure 1c,Figure 1d.
The model relies on several simplifying assumptions. We first assume that, before the two lamellae meet, the early spreading of each drop is unaffected by the presence of the other drop. The radial lamella velocity and the rim motion are therefore taken directly from the single-drop problem in §II. We also neglect the finite time required for the two drops to coalesce. When the two expanding rims first meet, the liquid contained in the individual rims is assumed to be incorporated instantaneously into the newly formed central sheet.
After collision, the two opposing radial fluxes meet at the symmetry plane. By symmetry, the velocity components normal to this plane, i.e. the components, are redirected upward by inertia, whereas the tangential components along the collision line, i.e. the components, are preserved [16, 23, 24]. This redirection forms a free-standing sheet in the – plane. Since the sheet is detached from the substrate except near its base, we neglect substrate friction within the sheet interior. Gravity is also neglected. For millimetric water drops impacting at , the Froude number is large, and the ratio of gravitational to inertial deceleration over the inertial–capillary time is over the range considered. Finally, after entering the free sheet, the flow is assumed to be inertia dominated. Each fluid parcel therefore follows a ballistic trajectory, with , while surface tension acts only through the bounding rim.
Dynamics and stages
Denote the maximum spreading radius of an isolated drop by . If , the two drops do not interact. If , the two spreading lamellae first meet at the midpoint at the collision onset time . For , the collision line lies in the symmetry plane and extends in the direction over , where . This quantity gives the lateral edge of the rising sheet, as sketched in Figure 1d. More generally, at a given position along the collision line, the local collision time is determined by .
At the collision line, the two incoming lamellae carry equal and opposite momentum in the direction. We assume that these normal components are redirected into the vertical direction, while the tangential components along are unchanged by the collision. At the point , with , the distance to either impact centre is . The velocity of the lamella coming from each parent drop can therefore be decomposed into components , where the signs correspond to the two drops centred at . After collision, the two contributions give the inlet conditions for the central sheet: , , and . Here and are the single-drop lamella velocity and thickness obtained in § II.
As the impact evolves, the boundary separating the drop and lamella regions reaches the collision line first at and then moves outward. In analogy with the local collision time, we define the local lamella-ending time by , with at . Thus both and depend on , and different portions of the sheet enter the subsequent stages at different times. We refer to the interval as the lamella-fed stage (L), and to as the post-lamella stage (PL).
In the PL stage, the thin lamella no longer supplies the sheet. The inlet is instead set by the bulk motion of the remaining liquid in the two drops, which continues to converge toward the symmetry plane after the lamella front has passed. This flow does not provide a closed-form inlet condition for the sheet, so we prescribe the PL inlet from simulations. Specifically, we use , , and . Here, we empirically propose is a polynomial in time, and with ; the constant is chosen to ensure continuity with the lamella-fed stage. These inlet forms are empirical, in the same spirit as the thickness function used to close the single-drop theory [1, 22]. The distinction is that is used as part of a general single-drop description, whereas a given PL inlet fit is valid only over the range of conditions from which it was obtained. The present fit covers all cases considered here; a substantially different parameter range would require a new calibration. Details of the extraction procedure and the range of validity are given in Appendix B.
For a fixed position , the local stage of the sheet is therefore determined by the two times and . The sheet is in the L stage when , and in the PL stage when , as shown in Figure 2a. At a given time , different parts of the same sheet can therefore be in different stages, as illustrated in Figure 2b. The inlet conditions for the sheet can then be written compactly as

Figure 2. (a) The two stages of the central sheet on the symmetry plane. In the lamella-fed stage (L, ) the parcel that reaches the apex is injected while the lamella still feeds the sheet, so the entire stall dynamics is contained in this stage; in the post-lamella stage (PL, ) the stall parcel is injected after feeding has ceased, and a two-stage integration is required. (b) The full three-dimensional sheet at an instant : points along the rim occupy different stages according to their lateral coordinate , separated by the boundary . (c) Regime map in the plane for water: the solid line bounds the region of no collision, and the dashed line separates the lamella-fed and post-lamella regimes.
It should be noted that and are analytically obtained as shown in Appendix A, while and are empirically obtained as shown in Appendix B.
Since the present study focuses on the rise of the central sheet, we define as the time at which the centreline height reaches its maximum value . Comparing with divides the – phase diagram into three regimes: no collision, L only, and L+PL, as shown in Figure 2c. The L-only regime occupies only a narrow part of the phase diagram, so the remainder of the paper focuses mainly on the sheet dynamics in the L+PL regime.
Semi-analytical model
Sheet domain
We first describe the velocity and thickness fields in the sheet interior, away from the bounding rim. The sheet is fed through the plane , where the values of , , and are prescribed by the stage-dependent conditions (1). After entering the free sheet, the fluid is assumed to experience neither viscous friction nor gravitational deceleration. Its velocity therefore remains constant along material paths, while the thickness evolves by mass conservation,
Here and . Consider a fluid element injected at time from , with inlet velocity . Since the velocity is constant along its trajectory, its position at a later time is
Expanding the mass-conservation equation then gives the thickness evolution along the same characteristic,
We now apply this construction to the L stage. On the inlet plane the conditions are those of (1), with the lamella velocity written as , so that
Applying the construction (3) and (4) with this inlet, the attenuation carried along each characteristic is its value at the origin, , and is therefore fixed once is fixed, since the velocity is conserved on material paths (2). The time dependence of is retained: it enters the final fields through , which varies from one characteristic to the next. At the inlet, the ratio of the vertical to horizontal velocity components is . Since both components remain constant along a characteristic, this ratio is preserved as the fluid element moves through the sheet. The characteristic starting from therefore satisfies
The corresponding injection time is obtained from the vertical displacement, which gives
Because the velocity is constant along each characteristic, . Substituting (6) and (7) into the inlet velocity then gives the L-stage velocity field
The thickness is then obtained from (4). From (8), we have and . Hence, along a characteristic, . On the same characteristic, , so and . Combining these relations gives
The bracketed quantity is therefore constant along each characteristic. At the injection point, , , and . Thus
The PL stage is treated in the same way, but with different inlet boundary conditions at . In this stage,
In this case the relation between and the injection variables is no longer algebraic. We therefore solve the inverse characteristic problem numerically. Once has been found, the thickness is obtained by integrating (4) along the same characteristic.
Rim domain
The sheet solution provides the velocity and thickness incident on the rim. We describe the rim by following material elements labelled by their source coordinate . The element labelled by has position , velocity , and cross-sectional diameter . The derivative is taken at fixed .
For a given , the direction of the incoming sheet flow is fixed by the impact geometry. We define
The incident sheet velocity and thickness are evaluated at the current rim position . The local flux incorporated into the rim element is then the sheet thickness multiplied by the relative velocity projected onto the incoming direction,
Here is defined for the rim element labelled by . Mass conservation gives
where is the cross-sectional area of the rim. Momentum conservation for the same element gives
The first term on the right-hand side is the momentum brought into the rim by the incorporated sheet flux. The second term is the capillary retraction force; its magnitude is because the rim bounds a free sheet with two liquid–air interfaces. These equations correspond to the single-drop rim balance (A8) with the free-rim geometric and capillary coefficients and . The rim position is advanced according to . In components, (14) becomes
The initial conditions are imposed when each rim element is created at the collision line, at . At this time,
The newly formed rim element contains the local sheet feed and the two single-drop rims that collide at the same location. Its initial speed is taken as the mass-weighted average
where and are the effective cross-sectional areas associated with the sheet feed and with each incoming single-drop rim, respectively. Projecting this speed onto the incoming direction gives the initial rim velocity,
Equations (12)–(18), driven by the sheet interior solution, form a closed system for the motion and growth of the rim. Integrating this system for each source coordinate gives the full side-view profile of the rising sheet in the plane, without any additional adjustable parameter in the rim model.
Comparison with experiment and simulation
The sheet fields of §III C 1 and the rim balances of §III C 2 together close the model. The sheet solution of §III C 1 is obtained separately and enters the rim problem as a known field, so the variables advanced in time are the rim position, velocity and cross-section, governed by (13)–(15) with , integrating them fixes the sheet profile and the apex height . We discretize the source coordinate into a fixed set of values along the collision line and integrate the rim system for each element, the value giving the apex. An element is created at its collision time with the initial state (16) and (18). At each step the sheet solution of §III C 1, driven by the stage-dependent inlet (1), is evaluated at each element’s current position to supply the incident velocity and thickness , each element drawing on the lamella-fed or post-lamella branch according to its own . The incident and drive the flux (12) and the balances (13)–(15), which are integrated by an explicit scheme at a fixed step . Halving the step and doubling the number of elements in leave the profiles below unchanged; the apex height is read from the element and the side-view profile from the element positions in the plane.
We now compare the predicted profiles with experiments and with direct numerical simulations (DNS). The DNS, described in Appendix C, provide an independent reference unaffected by the optical occlusion of the experimental side view, in which the central sheet can be partially hidden by the surrounding drop bulk. We first vary the Weber number at fixed half-spacing. Figure 3 compares the predicted profiles with the experiments of Goswami and Hardalupas [18] and with our simulations for four Weber numbers and four post-impact times at . The semi-analytical solution captures the overall sheet shape in the – plane, the growth of both the apex height and the basal footprint, and the increasing steepness of the dome as increases. The agreement is close at early times for all four Weber numbers. At the latest time, , the high- sheets have become unstable and pinch off in the simulations. The present unbroken-sheet model cannot represent this detached state, so the predicted and simulated profiles depart in this regime. This breakup sets the high- limit of the inertial rising-sheet description and is addressed in § V.
Figure 3. Side-view profiles of the central sheet at four Weber numbers (rows) and four post-impact times (columns), at fixed and . Red lines are the present theory, obtained by integrating the rim system (15); blue lines are the experiments of Goswami and Hardalupas [18] and green lines the present simulations.
We next examine the effect of the half-spacing at fixed . Figure 4 compares the predicted profiles with the simulations and with the experiments of Goswami and Hardalupas [18]. At this lower Weber number, the central sheet is relatively shallow, and the experimental side view is often partially obscured by the drop bulk. The comparison therefore relies mainly on the simulations, while the experimental contour is shown wherever it can be extracted reliably. The predicted profiles agree closely with the simulations across all four spacings and with the experimental contours where available.
Figure 4. Side-view profiles of the central sheet at four half-spacings (rows) and fixed , with the post-impact time annotated in each panel. Red lines are the present theory, obtained by integrating the rim system (15); blue lines are the experiments of Goswami and Hardalupas [18] and green lines the present simulations.
Gravity is included in the experiments and simulations but neglected in the semi-analytical model. As a result, the late-time descent of the sheet seen in the data is not reproduced. During the rising phase, however, the predicted profiles remain close to the measured and simulated contours. Together, the two comparisons span the principal control parameters, and , for water at . A further test at higher viscosity, , corresponding to a 40% glycerol–water mixture and lying slightly outside the low-viscosity range assumed in the derivation, is reported in Appendix § 0.10; close agreement is again obtained.
These comparisons show that the semi-analytical solution reproduces the full side-view shape over the parameter range of interest. In many applications, however, the main quantity of interest is the apex height along the symmetry plane , which is also the principal observable reported in experiments. We therefore next reduce the model to the centreline rim trajectory. On this centreline, the sheet fields simplify to closed-form expressions, and the condition for the maximum height reduces to a single algebraic relation.
Maximum apex height on the centreline
Lamella-fed stage
On the symmetry plane , the formulation of § III reduces to a one-dimensional problem. The lateral motion vanishes, , the rim normal is aligned with the direction, and . The vertical velocity field is therefore inherited directly from the lamella-fed solution, , with .
This vertical field determines the stretching of material elements in the direction. The centreline reduction must also retain the thinning caused by lateral stretching in the direction. At the base of the sheet, this lateral stretching rate is . In the full three-dimensional field it varies weakly with height as . For the centreline model, we approximate it by its base value, . This approximation keeps the solution algebraic while remaining close to the full three-dimensional rate over the range of heights relevant here. The centreline continuity equation is then
where the last term represents thinning by lateral stretching. Along the characteristics , this gives . Hence is conserved along each characteristic, and
with is the thickness of the sheet at the base . The exponent contains a unit contribution from vertical stretching and a contribution from lateral divergence. It is the centreline analogue of the horizontal invariant in the single-drop lamella solution of Gordillo et al. [1].
With and the rim normal along , the rim balances of § III reduce to
with . Integrating these equations gives the apex trajectory .
For the maximum height, the full trajectory can be reduced to a local balance at the apex. At the instant when the rim reaches its maximum height, . We approximate this instant by a quasi-steady stall condition, in which the upward momentum flux supplied by the sheet balances the capillary retraction of the rim. This gives
which we refer to as the stall condition throughout. Writing , the apex speed is , and the characteristic reaching at left the substrate at the injection time , with base thickness
where is the potential-flow thickness function of § 0.2. Substituting into the stall condition gives
This relation links the maximum height to the time . The time is obtained from the centreline rim integration, while the remaining quantities in the stall condition are explicit. Thus the prediction is semi-analytical: the centreline fields and the stall balance are closed in algebraic form, with only the weak dependence through $h_a evaluated numerically. If were constant, would cancel and the height would obey the explicit scaling
recovering in the inviscid limit . Over the present range of and , however, the variation of is not negligible. We therefore retain the full relation (24) and use (25) only as an asymptotic guide.
Post-lamella stage
Once the injection time of the apex parcel exceeds , the sheet is no longer supplied by the lamella. The centreline reduction of the previous subsection still applies, but the inlet conditions are now those of the post-lamella source. On the centreline, the boundary conditions (1) reduce to and . This source does not lead to a closed field of the form (20). Nevertheless, the apex still evolves along the vertical generator , and the characteristic construction can be carried out on this line.
A parcel released at time rises ballistically with the velocity acquired at injection, where we assume velocity has the form of based on the full simulation results as shown in Appendix B. Its position is therefore . Eliminating gives the vertical stretching rate , compared with in the lamella-fed stage. The lateral divergence also differs from the lamella-fed case. It is now set by the source value , rather than by the constant rate . Integrating (4) along the vertical generator gives
where the rational prefactor accounts for vertical stretching and the exponential factor accounts for lateral thinning. Thus, unlike the lamella-fed result (20), the post-lamella thickness does not reduce to a simple power law.
The maximum height is again estimated from the stall condition. The apex parcel travels from its injection time to the time of maximum height with the constant velocity acquired at injection. We define the flight-time ratio
so that . The corresponding apex height is
Evaluating the post-lamella field (26) at the apex gives the thickness of the sheet arriving at the rim
and the incoming apex speed is . Substitution into the stall condition (22) gives
Equation (30) determines the flight-time ratio implicitly. For a given Weber number, is found numerically, with supplied by the centreline rim integration as in the lamella-fed stage; the maximum height then follows from (28). In contrast to the lamella-fed stage, the post-lamella stage does not yield a closed power-law dependence of on . Both the vertical-stretching prefactor and the lateral-thinning exponential vary over the present parameter range, and the Weber-number dependence enters implicitly through . This behaviour is consistent with the absence of a single robust exponent for the apex height in the available experiments and simulations.
Comparison with experiment and simulation
Together, the lamella-fed and post-lamella reductions of § 0.4 A–IV B predict the apex height over the full range of impact conditions. The maximum height is only a scalar summary of the impact, however, and a more stringent test is whether the model also reproduces the time-resolved rise and the cross-sectional structure of the sheet. We carry out this test at , the value used by both Goswami and Hardalupas [18] and Zhang et al. [20]. Figure 5a–Figure 5c compares the predicted with the measurements of Goswami and Hardalupas [18] at , 52 and 64; the predicted curves, obtained by integrating the rim system (21) in time, reproduce the rise, the stall, and the approach to at all three Weber numbers without adjustable parameters. Figure 5d shows the predicted cross-section at for to 5, tracing the sheet as it narrows and climbs. Figure 5e–Figure 5g overlays this profile on the DNS field at , 3.0 and 4.0. At and 3.0 the agreement is excellent: the model captures not only the apex height but the full shape of the sheet, a continuous slender column tapering from the broad base to the rim. By , however, the apex has pinched off in the simulation, shedding a detached droplet above a base column that has stalled below it. The unbroken-sheet description, which assumes a continuous column from base to apex, cannot represent this configuration. The same detachment is observed in both experiment and simulation as increases, and marks the limit of the present model: the central sheet eventually breaks up under capillary action, the regime we turn to next.
Figure 5. All results in this figure correspond to water () at . (a–c) Time evolution of the sheet apex height for , 52 and 64; symbols denote the experimental measurements of Goswami and Hardalupas [18] and solid curves the present model, obtained by integrating the rim system (21) in time. (d) Predicted cross-sectional profiles of the central rising sheet at and successive times to 5, colour-coded by the horizontal colour bar. (e–g) Comparison between the predicted profile (curve) and direct numerical simulation (blue shaded region) at , 3.0 and 4.0, respectively. The rim has pinched off in (g).
Capillary pinch-off and the limit of the model
Capillary breakup and the apex cut-off
To capture this breakup, we extend the model with a capillary cut-off built on the thinning sheet itself. As the sheet rises it thins, and at high the post-lamella stage draws it out into a slender jet that is unstable to capillary perturbations; in the lamella-fed stage the sheet is still thick, fed, and rim-bounded, and no such instability develops. The analysis therefore proceeds along the post-lamella characteristics, on which the thickness is already known in closed form from (26), so a linear Rayleigh–Plateau treatment of the jet introduces no further parameters. The growth of these perturbations sets a breakup time, beyond which the detached fragments carry vertical momentum out of the sheet and the apex can rise no higher; the resulting height caps the inertial prediction at high .
We follow the parcel injected on the axis at along its characteristic; the local jet radius is , decreasing monotonically with . Because the thinning rate stays well below the capillary growth rate wherever breakup is binding, the most-amplifying mode at each instant is set by the local radius, and a quasi-steady dispersion relation applies pointwise along the characteristic. Taking the viscous Rayleigh relation [25, 26, 27] at the inviscid most-unstable wavenumber ,
with the viscous damping and the inviscid Rayleigh growth; is the classical Rayleigh [25] maximum growth rate and its viscous counterpart. Following Weber [28], breakup occurs when a perturbation grows from an initial amplitude to one comparable with the jet radius,
with , the standard value for capillary jet breakup and within the range 8–15 reported in the literature. Equation (32) sets the breakup time , and the parcel trajectory then fixes the breakup-limited apex height,
Every quantity on the right of (31)–(33) is either an inlet condition, an impact parameter (, ), or the literature threshold; none is adjusted to the comparisons below.
Comparison and range of validity
We first follow a single liquid (water, ) to high , beyond the experimentally accessible range. Figure 6a compares the inertial prediction (30) (solid), its lamella-fed asymptote (25) (triangle), the breakup-limited estimate (33) (dotted), the experiments of Goswami and Hardalupas [18] (squares), and the present DNS (diamonds). The data interpolate smoothly between the two predictions. At lower the apex jet, if it forms at all, breaks only after the sheet has essentially reached its maximum; little vertical momentum is lost, and follows the inertial prediction (30), which we denote as the height the sheet would attain in the absence of breakup. As rises the jet is drawn out earlier and thinner, breakup sets in during the rise, and the successive shed drops drain momentum from the apex, so that peels away from toward . Figure 6b sketches the two regimes observed in our simulations: at lower breakup is late and releases at most one drop, so ; at higher the jet pinches repeatedly during the rise, leaving well below .
Figure 6. Maximum central-sheet height versus Weber number at and . (a) The solid line is the inertial prediction (30) and the dotted line the breakup-limited height (33). Grey squares are the experiments of Goswami and Hardalupas [18] and blue diamonds the present DNS. (b) Schematic of the two regimes; the dashed extension marks the inertial height the sheet would reach without breakup. At lower (top) the sheet reaches before pinch-off; at higher (bottom) it pinches during the rise and stalls at .
Within the lower- range for which the model is built, pinch-off thus has only a modest effect on . To confirm that the inertial prediction holds across this range as and vary, we compare it against experiments and simulations over the full plane. In Figure 7a the grey surface is the inertial prediction, computed on a grid of 1600 points spanning the plane; the orange curve is its intersection with the contour, identical to the model curve of Figure 6a. Diamonds (present simulation) and squares (experiments of Goswami and Hardalupas [18, 19]) follow the colour convention of the ensemble, and red-edged circles flag , outside the low-viscosity range of the derivation. Restricting the comparison to , where the model assumptions hold, the prediction agrees with both experiment and simulation to within about (Figure 7b).
Figure 7. Maximum central-sheet height over the plane at . (a) The grey surface is the inertial prediction (30); the orange curve is its intersection with the iso- contour at (water), identical to the model curve in Figure 6a. Diamonds are the present simulation and squares the experiments of Goswami and Hardalupas [18, 19]; filled circles are the simulation results of Zhang et al. [20], coloured by (colour bar).Red-edged circles mark , outside the low-viscosity range of the derivation. (b) The same comparison restricted to , where the model assumptions hold; the blue band is the prediction across this range, coloured by , and symbols follow (a).
This agreement, however, holds only within the parameter range over which the derivation is valid. At high two distinct effects appear. Capillary breakup of the rising sheet is already accounted for through ; the single-drop lamella, however, may itself splash before reaching the collision line [8], corrupting the inlet upstream of the rim collision in a way the present description does not capture. At low , the inertial derivation rests on a thin viscous correction to a free-slip lamella, which loses accuracy as approaches and the boundary layer occupies an appreciable fraction of the lamella thickness. Between these limits the theory gives a self-consistent quantitative account of the central-sheet dynamics, from inertial feeding and two-directional thinning to capillary-limited expansion and the Rayleigh–Plateau closure, with no parameter adjusted beyond the once-calibrated post-lamella inlet.
Conclusions
We have developed a semi-analytical framework for the central rising sheet produced by the simultaneous impact of two drops on a solid substrate. The model couples the thin-film flow in the sheet to the bounding rim. It gives the velocity and thickness fields in the sheet and, through the centreline rim dynamics, the apex trajectory .
The framework describes the sheet through two feeding stages. During the lamella-fed stage, , the inlet flow is inherited from the single-drop lamella solution. The simple radial form of this inlet allows the ballistic characteristic construction to be carried out analytically. On the centreline, the vertical stretching and lateral divergence combine to give the conserved quantity , and the maximum height is reduced to an algebraic balance between inertial feeding and capillary retraction. The exponent reflects the combined thinning in the vertical and lateral directions. After , the lamella no longer reaches the collision line. The sheet is then supplied by the remaining bulk motion of the two drops, and the inlet no longer has a closed analytical form. We therefore prescribe the post-lamella inlet from simulations and continue the same characteristic construction based on the empirical boundary conditions. With this inlet, the model predicts the sheet trajectory and shape in good agreement with experiments and simulations over the parameter range considered.
The model also identifies the high- limit of the unbroken-sheet description. At sufficiently large Weber number, the apex thins into a slender jet and pinches off before the inertial sheet would reach its predicted maximum height. We account for this cutoff using a linear Rayleigh–Plateau analysis, with the time-dependent jet diameter and deceleration supplied by the sheet model. The resulting upper bound agrees well with both experiments and simulations.
The main limitation of the framework lies in the post-lamella inlet. In the lamella-fed stage, the inlet is determined by the single-drop lamella solution, and the injection point, injection time, velocity field, and thickness field can all be obtained explicitly. In the post-lamella stage, by contrast, the inlet is set by the residual bulk motion after the lamella front has passed the collision line. This flow does not reduce to a simple radial inlet law, and the inverse characteristic map cannot be written in closed form. The fitted post-lamella inlet used here captures the dynamics over the full range of conditions considered in this study, but it should not be regarded as a universal relation. A substantially different parameter range would require a new calibration.
This limitation concerns the inlet to the sheet rather than the subsequent sheet dynamics. Once the inlet velocity and thickness are specified, the rest of the model is deterministic: fluid parcels follow ballistic characteristics, the thickness evolves by mass conservation, the rim motion follows from mass and momentum balances, and the breakup cutoff is determined by the Rayleigh–Plateau criterion. The framework therefore separates the part of the problem that can be treated analytically from the single empirical input required in the post-lamella stage.
Although we have considered the simultaneous impact of two identical drops, the same construction can be extended to more general configurations. Natural next steps include nonsimultaneous impacts and unequal drop sizes, which are closer to the conditions encountered in spray cooling, coating, and inkjet printing.
Supplementary movies. Movies comparing the semi-analytical predictions with the numerical simulations over a range of parameters are provided as supplementary material.
Acknowledgements. S.S.H. thanks DeepSeek for assistance with spell-checking and grammar refinement of the manuscript. N.H. thanks Tachin Ruangkriengsin for the insightful discussion.
Funding. L.W.F. acknowledges the support from the National Natural Science Foundation of China (Grant No. 52276088).
Declaration of interests. The authors report no conflict of interest.
Lamella dynamics from a single impacting drop
Within the lamella, the depth-averaged radial velocity and film thickness satisfy the thin-film conservation laws
with [1]. The right-hand side of (A2) represents wall friction from the substrate boundary layer, whose thickness grows as in the stagnation-point regime [6, 7]. At the inner boundary the lamella is fed by the collapsing drop with the boundary conditions
where is a known function, independent of and , tabulated as a ninth-order polynomial fit to the potential-flow solution [1].
Since , we expand the velocity as . At leading order, the friction term is negligible, so remains constant along each characteristic. Let denote the time at which a fluid parcel enters the lamella. At that instant, the inner edge of the lamella is located at , and the boundary condition gives the parcel velocity . Because remains constant along the parcel trajectory, the parcel position at a later time satisfies . Solving for gives , and therefore
This self-similar velocity field is a purely kinematic consequence of the boundary condition (A3), independent of , or the details of . The leading-order continuity equation yields the invariant , giving the thickness profile based on as follows:
The first-order viscous correction, derived in full by Gordillo et al. [1], is
with . It is convenient to define a viscous attenuation factor by writing
so that measures the fraction of the inviscid leading-order velocity that survives viscous retardation. Because depends on position and time only through the combination , it is constant along each leading-order characteristic and varies slowly across the lamella at any instant. Over the parameter range considered it depends mainly on and only weakly on , changing by less than about 10% across the experimental range at fixed . We therefore treat as a constant, denoted , read off from the numerical solution for each and used as a fixed input to the lamella-fed stage in §??; the post-lamella stage instead takes its inlet directly from simulation, as described there.
The rim at advances at a velocity of and is governed by the mass and momentum balance equations
where and depend on the substrate wettability [1]. For a hydrophilic substrate and , with being the advancing contact angle. Eqns. (34)–(41), together with the initial conditions at the ejection time specified by Riboux and Gordillo [8, 29], form the complete single-drop problem. The rim ODE is integrated numerically; the lamella fields upstream of the rim are evaluated from the validated expressions (A4)–(A6). The resulting numerical solutions provide the time-dependent rim position together with the lamella state at arbitrary , both of which serve as input for the pair-impact analysis that follows.
Boundary conditions of the post-lamella stage
Once the lamella no longer feeds the sheet, the inflow is set by the bulk converging beneath the colliding rims, a flow that has no closed-form description and presents the same obstruction that has long kept the single-drop spreading problem open. We therefore fix the post-lamella inlet from the simulations, following the same route by which Gordillo et al. [1] close their lamella solution with a thickness function read from the numerical potential-flow solution. We extract the inlet thickness and injection velocity from the simulations and represent them by simple functions of time.
Both quantities are measured along the collision line. The rising sheet stands on the converging bulk beneath the rims, and the two meet at a sharp junction where the liquid narrows from the wide bulk below into the thin sheet, which we take as the base of the sheet. The inlet thickness is the width of the sheet at this junction and the injection velocity the mean vertical velocity across it, each averaged with the local liquid fraction. The same construction is applied to every case and every instant, so that and are obtained on a common footing and need no case-by-case adjustment.
Figure 8 shows the extracted and . Both vary only weakly between cases and show no systematic trend with or . The velocity roughly follows the decaying power law with taken to be the same for every case, the prefactor being set by continuity with the lamella-fed velocity at rather than by fitting. The thickness follows a cubic in time, again common to all cases, denoted in (1) and given by
Figure 8. Post-lamella inlet for four representative cases ( at , and at ). (a) Injection velocity (black, left axis) and thickness (blue, right axis); the red and green curves are the fitted forms and . (b) Injection flux . Line styles denote the four cases.
Although and each vary somewhat between cases, their product does not, because a thicker inlet is accompanied by a slower injection and the two variations cancel. It is this product, and not or separately, that fixes the momentum delivered to the rim, so the downstream solution depends on the inlet only through and is largely insensitive to the case-to-case differences in the two factors. This robustness comes with a corresponding limitation. The inlet is obtained by fitting, as indeed is the thickness function of the single-drop solution, but the two differ in range: is independent of and and so holds at any impact condition, whereas a fitted inlet holds over a finite window of conditions and is not universal. The present fit covers the full range examined in this work, and a sufficiently different regime would require its own.
Direct numerical simulations
The analytical model developed in the main text relies on a closed-form description of the lamella feed, whose validity over the impact-parameter range of interest is best assessed against three-dimensional simulations of the full Navier–Stokes equations. We carry out such simulations in Basilisk [30, 31], the same open-source platform used by Zhang et al. [20] for the simultaneous pair-drop impact problem, and adopt their numerical setup essentially without modification.
The two phases are governed by the incompressible Navier–Stokes equations with surface tension, in the dimensionless form introduced in § 0.2 (lengths scaled by , velocities by , time by , and pressure by ),
where is the full velocity field, the pressure, and the curvature and unit normal of the interface, and the interfacial Dirac distribution; , and are the groups defined in § 0.2 and § III, and is the upward unit vector. The dimensionless density and viscosity follow the volume fraction as and , with in the liquid, in the gas, and ratios , for water in air. Gravity is retained in the simulations through the body-force term; as shown in § III it is dynamically negligible over the range considered (), so it does not materially affect the sheet dynamics.
The equations are solved with an octree-based adaptive mesh refinement (AMR) method and a momentum-conserving geometric volume-of-fluid (VOF) formulation; the Bell–Colella–Glaz scheme is used for advection, viscous terms are treated implicitly, and surface tension is applied through the continuum-surface-force model [30, 32]. We use a cubic domain of side 16, with two identical drops of radius 1 released a small distance above the substrate and given the impact velocity directed downward, at a centre-to-centre horizontal separation , where is the impact parameter of § III. The substrate at is a no-slip wall with a static contact angle , imposed through a Neumann condition on the volume fraction, while the lateral and upper boundaries use outflow conditions [33, 34].
The maximum refinement level is , giving a minimum cell size , about 250 cells across the drop radius; refinement is driven by wavelet thresholds on the volume fraction and on the velocity, which keep the central sheet and rim on the finest grid throughout the rising phase. We verified that the apex height is grid-converged at this resolution, consistent with the mesh-convergence tests of Zhang et al. [20] at the same level. The setup reproduces the experiments of Goswami and Hardalupas [18] across the range examined here; the present simulations at shown in Figure 7(a) likewise recover the same experimental dataset.
Additional validation cases
The main-text comparisons validate the model against water across Weber number and drop spacing. To probe the remaining parameter direction, viscosity, we test it here against the 40% glycerol–water dataset of Goswami and Hardalupas [19], for which . This value lies marginally above the low-viscosity range over which our governing equations are derived, so the comparison probes the model just beyond its formal regime of validity. The remaining mixtures in that dataset, with glycerol fractions up to 80%, lie further outside the range of the single-drop spreading solution of Gordillo et al. [1] on which our framework rests; Goswami and Hardalupas [19] extend the spreading description to those cases through adjusted coefficients, but we do not pursue that extension here.
Figure 9a,b compares the side-view profiles at for water and the glycerol–water mixture at , 52 and 40.5. The added viscosity visibly reshapes the sheet: at a given Weber number the apex is markedly lower than for water, reflecting the enhanced viscous dissipation during the rise. The predicted contours nonetheless track the measured ones in both rows, so the model captures this viscous reshaping without any retuning of its parameters. Figure 9c–Figure 9e follows the apex height for the mixture across the full rise. The model reproduces the rate of rise at all three Weber numbers and the maximum height at ; at the two lower Weber numbers it slightly underpredicts the late-time apex, as expected when exceeds the range over which the lamella description is strictly valid.
Figure 9. Validation against the glycerol–water dataset of Goswami and Hardalupas [19] (). () Side-view profiles at for water () and the 40% glycerol–water mixture () at (columns). (–) Apex height for the mixture at the same Weber numbers. Filled contours and open circles are the experiments of Goswami and Hardalupas [19] (blue for water, grey for the mixture); red lines are the present theory.
Taken together, these comparisons show that the model continues to capture the central-sheet evolution for an Ohnesorge number modestly above its formal bound, the discrepancy remaining small and confined to the late stage of the rise. This supports its use across the full range of the main text.
References
References
- [1]J. M. Gordillo, G. Riboux, and E. S. Quintero, A theory on the spreading of impacting droplets, Journal of Fluid Mechanics 866, 298–315 (2019).DOI
- [2]A. Yarin, Drop impact dynamics: Splashing, spreading, receding, bouncing. . ., Annual Review of Fluid Mechanics 38, 159–192 (2006).DOI
- [3]C. Josserand and S. Thoroddsen, Drop impact on a solid surface, Annual Review of Fluid Mechanics 48, 365–391 (2016).DOI
- [4]X. Cheng, T.-P. Sun, and L. Gordillo, Drop impact dynamics: Impact force and stress distributions, Annual Review of Fluid Mechanics 54, 57–81 (2022).DOI
- [5]C. Clanet, C. Béguin, D. Richard, and D. Quéré, Maximal deformation of an impacting drop, Journal of Fluid Mechanics 517, 199 (2004).DOI
- [6]I. V. Roisman, Inertia dominated drop collisions. ii. an analytical solution of the navier–stokes equations for a spreading viscous film, Physics of Fluids 21, 10.1063/1.3129283 (2009).DOI
- [7]J. Eggers, M. A. Fontelos, C. Josserand, and S. Zaleski, Drop dynamics after impact on a solid wall: Theory and simulations, Physics of Fluids 22, 10.1063/1.3432498 (2010).DOI
- [8]G. Riboux and J. M. Gordillo, Experiments of drops impacting a smooth solid surface: A model of the critical impact speed for drop splashing, Physical Review Letters 113, 10.1103/physrevlett.113.024507 (2014).pubmed.ncbi.nlm.nih.gov/25062193
- [9]S. Wildeman, C. W. Visser, C. Sun, and D. Lohse, On the spreading of impacting drops, Journal of Fluid Mechanics 805, 636–655 (2016).DOI
- [10]V. Sanjay and D. Lohse, Unifying theory of scaling in drop impact: Forces and maximum spreading diameter, Physical Review Letters 134, 10.1103/physrevlett.134.104003 (2025).DOI
- [11]G. Liang and I. Mudawar, Review of mass and momentum interactions during drop impact on a liquid film, International Journal of Heat and Mass Transfer 101, 577–599 (2016).DOI
- [12]A. L. Yarin, I. V. Roisman, and C. Tropea, Collision Phenomena in Liquids and Solids (Cambridge University Press, Cambridge, 2017).DOI
- [13]J. Breitenbach, I. V. Roisman, and C. Tropea, From drop impact physics to spray cooling models: a critical review, Experiments in Fluids 59, 10.1007/s00348-018-2514-3 (2018).DOI
- [14]A. Moreira, A. Moita, and M. P. Panão, Advances and challenges in explaining fuel spray impingement: How much of single droplet impact research is useful?, Progress in Energy and Combustion Science 36, 554–580 (2010).DOI
- [15]H. A. Barnes, Y. Hardalupas, A. M. K. P. Taylor, and J. H. Wilkins, An investigation of the interaction between two adjacent impinging droplets, in Proceedings of the 15th International Conference on Liquid Atomisation and Spray Systems (ILASS), edited by G. Lavergne (ONERA, Toulouse, France, 1999) pp. 1–7.
- [16]I. Roisman, B. Prunet-Foch, C. Tropea, and M. Vignes-Adler, Multiple drop impact onto a dry solid substrate, Journal of Colloid and Interface Science 256, 396–410 (2002).DOI
- [17]N. E. Ersoy and M. Eslamian, Central uprising sheet in simultaneous and near-simultaneous impact of two high kinetic energy droplets onto dry surface and thin liquid film, Physics of Fluids 32, 10.1063/1.5135029 (2020).DOI
- [18]A. Goswami and Y. Hardalupas, Simultaneous impact of droplet pairs on solid surfaces, Journal of Fluid Mechanics 961, 10.1017/jfm.2023.249 (2023).DOI
- [19]A. Goswami and Y. Hardalupas, On the role of liquid viscosity during droplet-pair impacts on solid surfaces, Journal of Fluid Mechanics 1033, 10.1017/jfm.2026.11437 (2026).DOI
- [20]Z. Zhang, A. A. Castrejon-Pita, and W. Mostert, Numerical simulations of simultaneous pair-drop impacts and their energetics (2026).arxiv.org/abs/2601.19835
- [21]H. Wagner, Über stoß- und gleitvorgänge an der oberfläche von flüssigkeiten, ZAMM - Journal of Applied Mathematics and Mechanics / Zeitschrift für Angewandte Mathematik und Mechanik 12, 193–215 (1932).DOI
- [22]G. Riboux and J. M. Gordillo, Maximum drop radius and critical weber number for splashing in the dynamical leidenfrost regime, Journal of Fluid Mechanics 803, 516–527 (2016).
- [23]D. Hasson and R. E. Peck, Thickness distribution in a sheet formed by impinging jets, AIChE Journal 10, 752–754 (1964).DOI
- [24]N. Bremond and E. Villermaux, Atomization by jet impact, Journal of Fluid Mechanics 549, 273–306 (2006).DOI
- [25]L. Rayleigh, On the instability of jets, Proceedings of the London Mathematical Society s1-10, 4–13 (1878).
- [26]S. Chandrasekhar, Hydrodynamic and Hydromagnetic Stability (Oxford University Press, 1961).
- [27]J. Eggers and E. Villermaux, Physics of liquid jets, Reports on Progress in Physics 71, 036601 (2008).DOI
- [28]C. Weber, Zum zerfall eines flüssigkeitsstrahles, ZAMM - Journal of Applied Mathematics and Mechanics / Zeitschrift für Angewandte Mathematik und Mechanik 11, 136–154 (1931).DOI
- [29]G. Riboux and J. M. Gordillo, The diameters and velocities of the droplets ejected after splashing, Journal of Fluid Mechanics 772, 630–648 (2015).DOI
- [30]S. Popinet, An accurate adaptive solver for surface-tension-driven interfacial flows, Journal of Computational Physics 228, 5838–5866 (2009).DOI
- [31]S. Popinet, Numerical models of surface tension, Annual Review of Fluid Mechanics 50, 49–75 (2018).DOI
- [32]J. Brackbill, D. Kothe, and C. Zemach, A continuum method for modeling surface tension, Journal of Computational Physics 100, 335–354 (1992).DOI
- [33]V. Sanjay, P. Chantelot, and D. Lohse, When does an impacting drop stop bouncing?, Journal of Fluid Mechanics 958, 10.1017/jfm.2023.55 (2023).DOI
- [34]V. Sanjay, S. Lakshman, P. Chantelot, J. H. Snoeijer, and D. Lohse, Drop impact on viscous liquid films, Journal of Fluid Mechanics 958, 10.1017/jfm.2023.13 (2023).DOI







